Inserting Rows with Core — SQLAlchemy 2.0.0b1 documentation

SQLAlchemy 1.4 / 2.0 Tutorial

This page is part of the SQLAlchemy 2.0 Tutorial.

Previous: Working with Data | Next: Selecting Rows with Core or ORM

Inserting Rows with Core

When using Core, a SQL INSERT statement is generated using the _sql.insert() function - this function generates a new instance of _sql.Insert which represents an INSERT statement in SQL, that adds new data into a table.

ORM Readers - The way that rows are INSERTed into the database from an ORM perspective makes use of object-centric APIs on the _orm.Session object known as the unit of work process, and is fairly different from the Core-only approach described here. The more ORM-focused sections later starting at Inserting Rows with the ORM subsequent to the Expression Language sections introduce this.

The insert() SQL Expression Construct

A simple example of _sql.Insert illustrating the target table and the VALUES clause at once:

>>> from sqlalchemy import insert
>>> stmt = insert(user_table).values(name='spongebob', fullname="Spongebob Squarepants")

The above stmt variable is an instance of _sql.Insert. Most SQL expressions can be stringified in place as a means to see the general form of what’s being produced:

>>> print(stmt)
{opensql}INSERT INTO user_account (name, fullname) VALUES (:name, :fullname)

The stringified form is created by producing a _engine.Compiled form of the object which includes a database-specific string SQL representation of the statement; we can acquire this object directly using the _sql.ClauseElement.compile() method:

>>> compiled = stmt.compile()

Our _sql.Insert construct is an example of a “parameterized” construct, illustrated previously at Sending Parameters; to view the name and fullname bound parameters, these are available from the _engine.Compiled construct as well:

>>> compiled.params
{'name': 'spongebob', 'fullname': 'Spongebob Squarepants'}

Executing the Statement

Invoking the statement we can INSERT a row into user_table. The INSERT SQL as well as the bundled parameters can be seen in the SQL logging:

>>> with engine.connect() as conn:
...     result = conn.execute(stmt)
...     conn.commit()
{opensql}BEGIN (implicit)
INSERT INTO user_account (name, fullname) VALUES (?, ?)
[...] ('spongebob', 'Spongebob Squarepants')
COMMIT

In its simple form above, the INSERT statement does not return any rows, and if only a single row is inserted, it will usually include the ability to return information about column-level default values that were generated during the INSERT of that row, most commonly an integer primary key value. In the above case the first row in a SQLite database will normally return 1 for the first integer primary key value, which we can acquire using the _engine.CursorResult.inserted_primary_key accessor:

>>> result.inserted_primary_key
(1,)

Tip

_engine.CursorResult.inserted_primary_key returns a tuple because a primary key may contain multiple columns. This is known as a composite primary key. The _engine.CursorResult.inserted_primary_key is intended to always contain the complete primary key of the record just inserted, not just a “cursor.lastrowid” kind of value, and is also intended to be populated regardless of whether or not “autoincrement” were used, hence to express a complete primary key it’s a tuple.

Changed in version 1.4.8: the tuple returned by _engine.CursorResult.inserted_primary_key is now a named tuple fulfilled by returning it as a _result.Row object.

INSERT usually generates the “values” clause automatically

The example above made use of the _sql.Insert.values() method to explicitly create the VALUES clause of the SQL INSERT statement. This method in fact has some variants that allow for special forms such as multiple rows in one statement and insertion of SQL expressions. However the usual way that _sql.Insert is used is such that the VALUES clause is generated automatically from the parameters passed to the _future.Connection.execute() method; below we INSERT two more rows to illustrate this:

>>> with engine.connect() as conn:
...     result = conn.execute(
...         insert(user_table),
...         [
...             {"name": "sandy", "fullname": "Sandy Cheeks"},
...             {"name": "patrick", "fullname": "Patrick Star"}
...         ]
...     )
...     conn.commit()
{opensql}BEGIN (implicit)
INSERT INTO user_account (name, fullname) VALUES (?, ?)
[...] (('sandy', 'Sandy Cheeks'), ('patrick', 'Patrick Star'))
COMMIT{stop}

The execution above features “executemany” form first illustrated at Sending Multiple Parameters, however unlike when using the _sql.text() construct, we didn’t have to spell out any SQL. By passing a dictionary or list of dictionaries to the _future.Connection.execute() method in conjunction with the _sql.Insert construct, the _future.Connection ensures that the column names which are passed will be expressed in the VALUES clause of the _sql.Insert construct automatically.

INSERT…FROM SELECT

The _sql.Insert construct can compose an INSERT that gets rows directly from a SELECT using the _sql.Insert.from_select() method:

>>> select_stmt = select(user_table.c.id, user_table.c.name + "@aol.com")
>>> insert_stmt = insert(address_table).from_select(
...     ["user_id", "email_address"], select_stmt
... )
>>> print(insert_stmt)
{opensql}INSERT INTO address (user_id, email_address)
SELECT user_account.id, user_account.name || :name_1 AS anon_1
FROM user_account

INSERT…RETURNING

The RETURNING clause for supported backends is used automatically in order to retrieve the last inserted primary key value as well as the values for server defaults. However the RETURNING clause may also be specified explicitly using the _sql.Insert.returning() method; in this case, the _engine.Result object that’s returned when the statement is executed has rows which can be fetched:

>>> insert_stmt = insert(address_table).returning(address_table.c.id, address_table.c.email_address)
>>> print(insert_stmt)
{opensql}INSERT INTO address (id, user_id, email_address)
VALUES (:id, :user_id, :email_address)
RETURNING address.id, address.email_address

It can also be combined with _sql.Insert.from_select(), as in the example below that builds upon the example stated in INSERT…FROM SELECT:

>>> select_stmt = select(user_table.c.id, user_table.c.name + "@aol.com")
>>> insert_stmt = insert(address_table).from_select(
...     ["user_id", "email_address"], select_stmt
... )
>>> print(insert_stmt.returning(address_table.c.id, address_table.c.email_address))
{opensql}INSERT INTO address (user_id, email_address)
SELECT user_account.id, user_account.name || :name_1 AS anon_1
FROM user_account RETURNING address.id, address.email_address

Tip

The RETURNING feature is also supported by UPDATE and DELETE statements, which will be introduced later in this tutorial. The RETURNING feature is generally 1 only supported for statement executions that use a single set of bound parameters; that is, it wont work with the “executemany” form introduced at Sending Multiple Parameters. Additionally, some dialects such as the Oracle dialect only allow RETURNING to return a single row overall, meaning it won’t work with “INSERT..FROM SELECT” nor will it work with multiple row _sql.Update or _sql.Delete forms.

1

There is internal support for the _postgresql.psycopg2 dialect to INSERT many rows at once and also support RETURNING, which is leveraged by the SQLAlchemy ORM. However this feature has not been generalized to all dialects and is not yet part of SQLAlchemy’s regular API.

See also

_sql.Insert - in the SQL Expression API documentation